Empirically Assessing Opportunities for Prefetching and Caching in Mobile Apps

Presentation slides of the paper "Empirically Assessing Opportunities for Prefetching and Caching in Mobile Apps" at the International Conference on Automated Software Engineering (ASE) 2018.

The presentation can be found at https://youtu.be/iT1pNy370ZM

Transcript

1. Empirically Assessing Opportunities for Prefetching and Caching in Mobile Apps Yixue Zhao, Paul Wat, Marcelo Schmitt Laser, and Nenad Medvidović University of Southern California September 6, 2018 @Montpellier
2. Mobile Apps Everywhere
3. Network Latency Latency!
4. Network Latency Wireless network characteristics Ø Low bandwidth Ø Unstable connectivity
5. Network Latency Performance Bottleneck Network Speed
6. Prefetching Caching
7. Prefetching & Caching local cache
8. Prefetching & Caching immediate response local cache ICSE 2018. Leveraging Program Analysis to Reduce User-Perceived Latency in Mobile Applications.
9. Browser Mobile App Browser vs. Mobile App Ø Lack of work Ø Large body of work ?
10. Browser Mobile App Browser vs. Mobile App Ø Lack of work Ø “Light” requests Ø Large body of work Ø “Rich” requests ?
11. Browser Mobile App Browser vs. Mobile App Ø Lack of work Ø “Light” requests Ø Single round-trip time Ø Large body of work Ø “Rich” requests Ø Loading multiple sub-resources ?
12. Browser Mobile App Browser vs. Mobile App Ø Lack of work Ø “Light” requests Ø Single round-trip time Ø Prefetch separate requests Ø Large body of work Ø “Rich” requests Ø Loading multiple sub-resources Ø Prefetch sub-resources ?
13. Conclusion Ø Browser à Mobile app Ø Lack of prefetching & caching in mobile apps Browser vs. Mobile App
14. Empirical Study Can we prefetch & cache in mobile apps?
15. Research Questions Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
16. GET requests Ø Read-only Ø No side effects Ø Safe to prefetch RQ: Prefetchability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
17. 1. What is the number of GET requests per app? 2. What is the percentage of GET requests among all HTTP requests? 3. How prevalent are GET requests across different app categories? RQ: Prefetchability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
18. RQ: Cacheability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
19. Prefetchable ≠ Cacheable Ø Cache staleness Ø Rely on Expires, Cache-Control Ø Trustworthy? RQ: Cacheability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
20. 4. How prevalent are Expires headers? 5. Are Expires headers trustworthy? 6. How prevalent are Cache-Control headers? 7. Are Cache-Control headers trustworthy? RQ: Cacheability Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
21. RQ: Redundancy Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
22. Redundant requests Ø Ostensibly redundant: same URL Ø Truly redundant: same URL & same response RQ: Redundancy Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests Cache me J
23. 8. How prevalent are redundant HTTP requests? 9. Are the identified ostensibly redundant requests truly redundant? RQ: Redundancy Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
24. Workflow
25. Workflow Overview Send Requests t=0, 10, 30, 60s #req ≥ 4 Instrumented Apps App Testing Monkey Exerciser download Manual Examination Initial Apps
26. Ø 1,687 top-ranked apps Ø 33 app categories e.g., Beauty, Tools Workflow: Download download Google Play Initial Apps 1,687
27. Ø Instrumented 1,308 apps to capture needed information by Soot Ø App size: 16 KB to 103.4 MB Ø #Request per app: 0 to 1,243 Workflow: Instrumentation download Google Play Initial Apps Instrumented Apps 1,308 1,687
28. Ø 3,000 random events by Monkey Ø NoxPlayer emulator Ø Wi-Fi network settings Workflow: Testing download Google Play Initial Apps Instrumented Apps App Testing Monkey Exerciser 1,687 1,308
29. Workflow: Inspection Ø Low network usage
30. Workflow: Inspection Ø Low network usage Ø Manual inspection Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps
31. Ø Low network usage Ø Manual inspection Ø 6 recurring reasons e.g. obfuscation Workflow: Inspection Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps
32. 1 ≤ #Requests ≤ 3 Ø Load config files Ø Log in with Facebook Ø Monitoring services Workflow: Inspection Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps
33. #req ≥ 4 Workflow: Final Apps Final apps (#req ≥ 4) Ø 451 apps Ø 33 app categories e.g. Beauty, Tools Ø #Req per app Avg: 35 Max: 1,243
34. Workflow Overview Send Requests t=0, 10, 30, 60s #req ≥ 4 Instrumented Apps App Testing Monkey Exerciser download Google Play Manual Examination Initial Apps
35. Findings
36. Findings Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
37. Prefetchability of HTTP requests Ø What is the number of GET requests per app? Min 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
38. Prefetchability of HTTP requests Ø What is the number of GET requests per app? Max 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
39. Prefetchability of HTTP requests Ø What is the number of GET requests per app? Avg 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
40. Prefetchability of HTTP requests Ø What is the number of GET requests per app? 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 1,243
41. Prefetchability of HTTP requests Ø What is the number of GET requests per app? 0 25 50 75 100 125 150 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 1,243 263 182 802 286 174 166
42. Prefetchability of HTTP requests Ø What is the percentage of GET requests per app? 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
43. Prefetchability of HTTP requests Ø GET is pervasive à prefetching opportunity 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
44. Prefetchability of HTTP requests Ø GET is pervasive à prefetching opportunity Ø Highest avg: Beauty 94% Ø Lowest avg: Dating 43% 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
45. Prefetchability of HTTP requests 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Beauty 94%
46. Prefetchability of HTTP requests 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Dating 43%
47. Findings Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
48. Cacheability of HTTP responses Ø Expires & Cache-Control response headers Ø Numbers Ø Percentages 0 5 10 15 20 25 30 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Percentages of Expires Numbers of Expires
49. Cacheability of HTTP responses Avg of Expires percentages Avg of Cache-Control percentages 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
50. Cacheability of HTTP responses Ø Avg: Expires 53%, Cache-Control 65% Avg of Cache-Control percentages 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Avg of Expires percentages
51. Cacheability of HTTP responses Ø Avg: Expires 53%, Cache-Control 65% Ø Both headers are not always included Avg of Cache-Control percentages 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 Avg of Expires percentages
52. Cacheability of HTTP responses Ø Are Expires and Cache-Control headers trustworthy?
53. Cacheability of HTTP responses Instrumented Apps App Testing Manual Examination Monkey Exerciser Send Requests t=0, 10, 30, 60s #req ≥ 4 download Google Play Initial Apps
54. Cacheability of HTTP responses Ø Send original request @ time(t) Ø Send same request @ time(t+x), x=10, 30, 60s t = original time x = time period after t
55. Cacheability of HTTP responses Ø Untrustworthy scenarios 1. time(exp) ≤ time(t) 2. time(t) < time(exp) ≤ time(t+x) ∧ response@(t) = response@(t+x) 3. time(exp) > time(t+x) ∧ response@(t) ≠ response@(t+x) t = original time x = time period after t
56. Cacheability of HTTP responses Ø Untrustworthy scenarios t = original time x = time period after t t t + x request@t 2018 Sep 1 2018 Sep 6
57. Cacheability of HTTP responses Ø Untrustworthy scenarios t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6
58. Cacheability of HTTP responses Ø Untrustworthy scenarios t = original time x = time period after t t t + x expire time 2018 Sep 1 2018 Sep 6
59. Cacheability of HTTP responses Ø 1. expire before request is sent t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 expire time: 1999 Jan 1
60. Cacheability of HTTP responses Ø 2. should expire but response is unchanged t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 expire time: 2018 Sep 3
61. Cacheability of HTTP responses Ø 3. should not expire but response is changed t = original time x = time period after t t t + x 2018 Sep 1 2018 Sep 6 expire time: 2018 Sep 10
62. Cacheability of HTTP responses Ø Avg trustworthy: Expires 25%, Cache-Control 77% (3x) Avg of trusted Expires Avg of trusted Cache-Control 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
63. Cacheability of HTTP responses Ø Avg trustworthy: Expires 25%, Cache-Control 77% (3x) Ø 0% and 100% apps Avg of trusted Expires Avg of trusted Cache-Control 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 0 25 50 75 100 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
64. Cacheability of HTTP responses Ø Should not depend on the headers completely
65. Cacheability of HTTP responses Ø Should not depend on the headers completely Ø No reliable alternatives
66. Cacheability of HTTP responses Ø Should not depend on the headers completely Ø No reliable alternatives Ø Research opportunities J
67. Findings Ø Prefetchability of HTTP requests Ø Cacheability of HTTP responses Ø Redundancy of HTTP requests
68. Redundancy of HTTP requests Ø Ostensibly redundant requests Ø Avg: 20% 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Percentages of ostensibly redundant requests
69. Redundancy of HTTP requests Ø Truly redundant requests Ø Avg: 92% 0 25 50 75 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Percentages of truly redundant requests
70. Redundancy of HTTP requests Ø Most redundant requests are truly redundant
71. Redundancy of HTTP requests Ø Most redundant requests are truly redundant Ø What does this tell us?
72. Implications for prefetching Ø “Avg” app Ø #GET request
    28
73. Implications for prefetching Ø “Avg” app Ø #GET request
    28 Ø Response time: 800ms Time saved: 22s
74. Implications for caching Ø “Avg” app Ø #GET request
    28 Ø Redundant requests: 20% Ø Truly redundant requests: 92% Ø #Cacheable request: 6 Time saved: 4s
75. Summary
76. Summary Empirical study to understand HTTP characteristics in mobile apps Ø Prefetchability of requests Ø Cacheability of responses Ø Redundancy of requests
77. Lessons learned Ø Opportunity for prefetching & caching Ø App-specific characteristics Ø Domain-specific characteristics Summary
78. Summary Contributions Ø Bridge the gap between browser and mobile app domain Ø Guidelines for developers Ø Motivate future research
79. Behind the Scene: co-authors Marcelo Schmitt Laser Paul Wat advisor: Nenad Medvidović
80. Behind the Scene: co-authors + sponsors Marcelo Schmitt Laser Paul Wat advisor: Nenad Medvidović
81. Contact: yixue.zhao@usc.edu Github: https://github.com/felicitia/PALOMA-Analysis/tree/empirical